Agent-Based Model for Surface Aerothermal Prediction Based on Ensemble Learning

This paper employs an ensemble learning algorithm to construct a surrogate model for predicting aerodynamic heat flux on hypersonic vehicle surfaces, based on the influence of flight state parameters on physical characteristic state parameters. First, an aerothermal dataset for the vehicle surface is obtained through computational fluid dynamics. Second, based on the strong nonlinear coupling relationship between flight state parameters and the physical characteristics of the flowfield, a set of physical characteristic state parameters was constructed as supplementary input features. Finally, the eXtreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) ensemble learning algorithms are separately adopted to develop point-to-point aerothermal surrogate models denoted as physical characteristic state parameters. Validation using a hypersonic double-ellipsoid model demonstrates that the proposed model achieves mean absolute percentage errors below 10% for the full-surface heat flux prediction, with the relative error at the stagnation point kept within 3%. In addition, XGBoost outperforms LightGBM in full-surface heat flux prediction accuracy. Meanwhile, the prediction of a single flight condition takes only about 2 s. This model effectively shortens the aerodynamic heat calculation cycle while meeting the accuracy requirements for aerodynamic heat prediction.